myagent-ai 1.11.2 → 1.12.0

package/config.py

@@ -100,6 +100,7 @@ class ModelEntry:
     input_modes: List[str] = field(default_factory=lambda: ["text"])  # 支持的输入模式: text, image, video, audio
     reasoning: bool = True       # 是否支持推理（如 o1 系列）
     enabled: bool = True
+    is_global_fallback: bool = True  # 是否作为全局兜底模型（保障系统运行的最终兜底模型）
 
 
 @dataclass

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "myagent-ai",
-  "version": "1.11.2",
+  "version": "1.12.0",
   "description": "本地桌面端执行型AI助手 - Open Interpreter 风格 | Local Desktop Execution-Oriented AI Assistant",
   "main": "main.py",
   "bin": {

package/web/api_server.py

@@ -2594,73 +2594,52 @@ class ApiServer:
         return web.json_response({"ok": True, "hot_reload": True})
 
     async def handle_test_llm(self, request):
-        """POST /api/llm/test - 测试 LLM 连接，支持使用请求体中的临时配置"""
-        print(f"handle_test_llm 被调用!", flush=True)
+        """POST /api/llm/test - 独立测试指定模型连接，不使用任何兜底/fallback逻辑"""
         try:
-            # 尝试从请求体读取临时配置
-            temp_client = None
-            try:
-                data = await request.json()
-                print(f"请求数据: {data}", flush=True)
-                api_key = data.get("api_key", "")
-                base_url = data.get("base_url", "")
-                model = data.get("model", "")
-                api_type = data.get("api_type", "")
-                temperature = data.get("temperature", 0.1)
-                # 如果请求中有 provider，直接使用；否则根据 api_type 或 base_url 推断
-                provider = data.get("provider", "")
-                if not provider:
-                    if api_type in ("openai-chat", "openai-completions"):
-                        provider = "openai"
-                    elif api_type == "anthropic":
-                        provider = "anthropic"
-                    elif api_type == "ollama":
-                        provider = "ollama"
-                    elif "zhipu" in base_url or "bigmodel" in base_url:
-                        provider = "zhipu"
-                    elif "anthropic" in base_url:
-                        provider = "anthropic"
-                    elif "ollama" in base_url or "localhost" in base_url or "127.0.0.1" in base_url:
-                        provider = "ollama"
-                    elif "dashscope" in base_url or "qwen" in base_url.lower() or "modelscope" in base_url:
-                        provider = "custom"  # DashScope/ModelScope 使用自定义 provider
-                    else:
-                        provider = "openai"
-                print(f"准备创建 temp_client: provider={provider}, api_key={api_key[:10] if api_key else 'empty'}...", flush=True)
-                if api_key or base_url or model:
-                    # 使用请求体中的配置创建临时客户端测试
-                    from core.llm import LLMClient
-                    temp_client = LLMClient(
-                        provider=provider,
-                        api_key=api_key or self.core.llm.api_key if self.core.llm else "",
-                        base_url=base_url or self.core.llm.base_url if self.core.llm else "",
-                        model=model or self.core.llm.model if self.core.llm else "gpt-4",
-                        temperature=temperature,
-                        timeout=30,
-                        max_retries=1,
-                    )
-                    print(f"temp_client 创建成功: {temp_client}, base_url={temp_client.base_url}", flush=True)
+            data = await request.json()
+            api_key = data.get("api_key", "")
+            base_url = data.get("base_url", "")
+            model = data.get("model", "")
+            api_type = data.get("api_type", "")
+            temperature = data.get("temperature", 0.1)
+            provider = data.get("provider", "")
+
+            if not model:
+                return web.json_response({"ok": False, "error": "缺少 model 参数"})
+
+            # 根据 api_type 或 base_url 推断 provider
+            if not provider:
+                if api_type == "anthropic":
+                    provider = "anthropic"
+                elif api_type == "ollama":
+                    provider = "ollama"
+                elif "zhipu" in base_url or "bigmodel" in base_url:
+                    provider = "zhipu"
+                elif "anthropic" in base_url:
+                    provider = "anthropic"
+                elif "ollama" in base_url or "localhost" in base_url or "127.0.0.1" in base_url:
+                    provider = "ollama"
+                elif "dashscope" in base_url or "qwen" in base_url.lower() or "modelscope" in base_url:
+                    provider = "custom"
                 else:
-                    print("条件不满足，temp_client 未创建", flush=True)
-            except Exception as e:
-                print(f"创建 temp_client 异常: {e}", flush=True)
-                logger.warning(f"读取请求体失败: {e}")
-                pass  # 如果没有请求体或解析失败，使用当前配置
-
-            client = temp_client or self.core.llm
-            print(f"DEBUG: temp_client={temp_client is not None}, client={client}, client_type={type(client).__name__}, client_base_url={getattr(client, 'base_url', 'N/A')}", flush=True)
-            if not client:
-                print("没有 LLM 客户端", flush=True)
-                return web.json_response({"ok": False, "error": "没有配置 LLM 客户端"})
-            msg = await client.chat([Message(role="user", content="Hi, reply OK")])
-            print(f"TEST: msg.success={msg.success}, msg.content={repr(msg.content[:50] if msg.content else '')}, msg.error={msg.error}", flush=True)
-            # 必须检查 msg.success，否则即使请求失败也会返回成功
+                    provider = "openai"
+
+            # 纯粹使用请求体中的参数创建临时客户端，绝不 fallback 到全局配置
+            from core.llm import LLMClient
+            temp_client = LLMClient(
+                provider=provider,
+                api_key=api_key,
+                base_url=base_url,
+                model=model,
+                temperature=temperature,
+                timeout=30,
+                max_retries=1,
+            )
+            msg = await temp_client.chat([Message(role="user", content="Hi, reply OK")])
             if not msg.success:
-                print(f"返回失败: {msg.error}", flush=True)
                 return web.json_response({"ok": False, "error": msg.error or "LLM 调用失败"})
             return web.json_response({"ok": True, "response": msg.content[:100] if msg.content else ""})
         except Exception as e:
-            print(f"handle_test_llm 异常: {e}", flush=True)
             return web.json_response({"ok": False, "error": str(e)})
 
     async def handle_llm_usage(self, request):
@@ -2680,6 +2659,7 @@ class ApiServer:
                 "input_modes": m.input_modes,
                 "reasoning": m.reasoning,
                 "enabled": m.enabled,
+                "is_global_fallback": getattr(m, 'is_global_fallback', True),
                 "has_api_key": bool(m.api_key),
             })
         return web.json_response(models)
@@ -2708,6 +2688,7 @@ class ApiServer:
             input_modes=data.get("input_modes", ["text"]),
             reasoning=data.get("reasoning", False),
             enabled=data.get("enabled", True),
+            is_global_fallback=data.get("is_global_fallback", True),
         )
         self.core.config.models_library.append(entry)
         self.core.config_mgr.save()
@@ -2721,7 +2702,7 @@ class ApiServer:
         found = False
         for m in self.core.config.models_library:
             if m.id == model_id:
-                for k in ("name", "provider", "api_type", "model", "base_url", "max_tokens", "temperature", "context_window", "input_modes", "reasoning", "enabled"):
+                for k in ("name", "provider", "api_type", "model", "base_url", "max_tokens", "temperature", "context_window", "input_modes", "reasoning", "enabled", "is_global_fallback"): 
                     if k in data:
                         setattr(m, k, data[k])
                 if data.get("api_key"):
@@ -3062,20 +3043,40 @@ class ApiServer:
                 "is_backup": False,
             })
 
-        # 如果没有找到任何主模型配置，使用全局默认 LLM 作为 fallback
+        # 如果没有找到任何主模型配置，使用模型库中标记为全局兜底的模型
         if not chain:
-            chain.append({
-                "id": "default",
-                "name": llm_defaults.model,
-                "provider": llm_defaults.provider,
-                "api_type": llm_defaults.api_type,
-                "model": llm_defaults.model,
-                "base_url": llm_defaults.base_url,
-                "api_key": llm_defaults.api_key,
-                "temperature": llm_defaults.temperature,
-                "max_tokens": llm_defaults.max_tokens,
-                "is_backup": False,
-            })
+            fallback_model = None
+            for me in self.core.config.models_library:
+                if getattr(me, 'is_global_fallback', True) and me.enabled:
+                    fallback_model = me
+                    break
+            if fallback_model:
+                chain.append({
+                    "id": fallback_model.id,
+                    "name": fallback_model.name,
+                    "provider": fallback_model.provider or llm_defaults.provider,
+                    "api_type": getattr(fallback_model, 'api_type', '') or llm_defaults.api_type,
+                    "model": fallback_model.model or fallback_model.id,
+                    "base_url": fallback_model.base_url or llm_defaults.base_url,
+                    "api_key": fallback_model.api_key or llm_defaults.api_key,
+                    "temperature": fallback_model.temperature if fallback_model.temperature is not None else llm_defaults.temperature,
+                    "max_tokens": fallback_model.max_tokens if fallback_model.max_tokens else llm_defaults.max_tokens,
+                    "is_backup": False,
+                })
+            else:
+                # 兜底到全局默认 LLM 配置
+                chain.append({
+                    "id": "default",
+                    "name": llm_defaults.model,
+                    "provider": llm_defaults.provider,
+                    "api_type": llm_defaults.api_type,
+                    "model": llm_defaults.model,
+                    "base_url": llm_defaults.base_url,
+                    "api_key": llm_defaults.api_key,
+                    "temperature": llm_defaults.temperature,
+                    "max_tokens": llm_defaults.max_tokens,
+                    "is_backup": False,
+                })
 
         # 备用模型列表
         backup_ids = agent_cfg.get("backup_model_ids", [])

package/web/ui/chat/chat_main.js

@@ -304,6 +304,7 @@ function initChat() {
   // Enable/disable send button
   document.getElementById('userInput').addEventListener('input', function() {
     document.getElementById('sendBtn').disabled = !this.value.trim();
+    saveDraft();
   });
   // Load task plan if in exec mode (collapsed by default)
   if (state.chatMode === 'exec') {
@@ -1182,6 +1183,7 @@ async function selectAgent(agentPath) {
   // 如果 loadSessions 已经 auto-selected 了 session，UI 已由 selectSession 设置好，不再覆盖
 
   document.getElementById('userInput').focus();
+  loadDraft();
   // Reload task plan if in exec mode
   if (state.chatMode === 'exec') loadTaskPlan();
   // Reset escalation and update exec mode UI
@@ -1668,6 +1670,7 @@ function newChat() {
   updateSidebarAgentIndicator();
   const userInput = document.getElementById('userInput');
   if (userInput) userInput.focus();
+  loadDraft();
   // 移动端：新建会话后自动收起侧边栏
   if (isMobile()) closeMobileSidebar();
 }
@@ -1804,6 +1807,7 @@ async function selectSession(id) {
     }
   } catch (_) {}
   document.getElementById('userInput').focus();
+  loadDraft();
   if (isMobile()) closeMobileSidebar();
 }
 

package/web/ui/chat/flow_engine.js

@@ -3,7 +3,36 @@
 // ── 负责消息发送、SSE 流式处理、大文本检测与分段、
 //    消息注入、执行进度轮询、输入处理等核心流程 ──
 // ══════════════════════════════════════════════════════
-//
+
+// ── 输入框草稿持久化（按 agent + session 维度保存，刷新页面不丢失） ──
+var _draftTimer = null;
+function _draftKey() {
+  return 'myagent_draft_' + (state.activeAgent || 'default') + '_' + (state.activeSessionId || '__new__');
+}
+function saveDraft() {
+  if (_draftTimer) clearTimeout(_draftTimer);
+  _draftTimer = setTimeout(function() {
+    var input = document.getElementById('userInput');
+    if (!input) return;
+    var val = input.value;
+    if (val) { localStorage.setItem(_draftKey(), val); }
+    else { localStorage.removeItem(_draftKey()); }
+  }, 300);
+}
+function loadDraft() {
+  var input = document.getElementById('userInput');
+  if (!input) return;
+  var val = localStorage.getItem(_draftKey());
+  input.value = val || '';
+  input.style.height = 'auto';
+  if (input.value) autoResize(input);
+  document.getElementById('sendBtn').disabled = !input.value.trim();
+}
+function clearDraft() {
+  localStorage.removeItem(_draftKey());
+}
+
+
 // 依赖的全局对象（由 chat_main.js 提供）：
 //   state, api(), toast(), escapeHtml(), renderMarkdown(),
 //   renderMessages(), scrollToBottom(), formatSessionName(),
@@ -1145,6 +1174,7 @@ async function sendMessage() {
   input.value = '';
   input.style.height = 'auto';
   document.getElementById('sendBtn').disabled = true;
+  clearDraft();
 
   // Show typing
   state.isGenerating = true;
@@ -1695,6 +1725,7 @@ async function handleInjectChoice(choice) {
   const input = document.getElementById('userInput');
   input.value = '';
   input.style.height = 'auto';
+  clearDraft();
 
   try {
     const rawSid = state.activeSessionId ? state.activeSessionId.replace(`${state.activeAgent}_`, '') : 'web_default';

package/web/ui/chat/groupchat.js

@@ -610,9 +610,7 @@ async function sendGroupChat() {
   input.value = '';
   input.style.height = 'auto';
   document.getElementById('sendBtn').disabled = true;
-
-  // Show typing
-  state.isGenerating = true;
+  clearDraft();
   showTypingIndicator();
   document.getElementById('sendBtn').style.display = 'none';
   document.getElementById('stopBtn').style.display = '';

package/web/ui/index.html

@@ -1120,47 +1120,10 @@ async function resetAgentPerms(name){
 
 // ========== LLM ==========
 async function renderLLM(){
-  const [c,u,models]=await Promise.all([api('/api/llm'),api('/api/llm/usage'),api('/api/models')]);
-  if(c.error){$('content').innerHTML='<div class="empty" style="color:var(--danger)">加载失败: '+escHtml(c.error)+'</div>';return}
+  const [u,models]=await Promise.all([api('/api/llm/usage'),api('/api/models')]);
   // 更新模型缓存
   allModelsCache=Array.isArray(models)?models:[];
-  const providerOpts=['openai','anthropic','ollama','zhipu','deepseek','moonshot','qwen','custom','modelscope'].map(p=>`<option value="${p}" ${c.provider===p?'selected':''}>${p}</option>`).join('');
-  const apiTypeOpts=['openai-completions','openai-chat','anthropic','ollama'].map(t=>`<option value="${t}" ${c.api_type===t?'selected':''}>${t}</option>`).join('');
-  const isCustomApiType=c.api_type&&!['openai-completions','openai-chat','anthropic','ollama'].includes(c.api_type);
-  const inputModesArr=c.input_modes||['text'];
-  const inputModesChecked=['text','image','video','audio'].map(m=>`<label style="display:inline-flex;align-items:center;gap:4px;margin-right:12px"><input type="checkbox" id="llmInputMode_${m}" value="${m}" ${inputModesArr.indexOf(m)>-1?'checked':''}>${m}</label>`).join('');
-  let html=`<div class="card"><h3>全局模型配置</h3>
-    <div class="form-row">
-      <div class="form-group"><label>Provider</label><select id="llmProvider">${providerOpts}</select></div>
-      <div class="form-group">
-        <label>API 类型</label>
-        <select id="llmApiTypeSelect" onchange="toggleApiTypeInput('llm')">
-          ${apiTypeOpts}
-          <option value="custom" ${isCustomApiType?'selected':''}>custom (自定义)</option>
-        </select>
-      </div>
-    </div>
-    <div id="llmApiTypeCustomWrap" style="${isCustomApiType?'':'display:none'}" class="form-group">
-      <label>自定义 API 类型</label>
-      <input id="llmApiTypeCustom" value="${isCustomApiType?escHtml(c.api_type):''}" placeholder="输入自定义 API 类型">
-    </div>
-    <div class="form-group"><label>模型</label><input id="llmModel" value="${c.model||''}" placeholder="如 gpt-4o"></div>
-    <div class="form-group"><label>Base URL</label><input id="llmUrl" value="${c.base_url||''}" placeholder="https://api.openai.com/v1"></div>
-    <div class="form-group"><label>API Key</label><input id="llmKey" type="password" value="" placeholder="${c.api_key_set?'已配置（留空不修改）':'未配置'}"></div>
-    <div class="form-row">
-      <div class="form-group"><label>上下文窗口</label><input id="llmContextWindow" type="number" value="${c.context_window||128000}" placeholder="128000"></div>
-      <div class="form-group"><label>最大输出</label><input id="llmMaxTokens" type="number" value="${c.max_tokens||4096}" placeholder="4096"></div>
-    </div>
-    <div class="form-row">
-      <div class="form-group"><label>Temperature</label><input id="llmTemp" type="number" step="0.1" value="${c.temperature||0.7}" placeholder="0.1"></div>
-      <div class="form-group" style="display:flex;align-items:center">
-        <input id="llmReasoning" type="checkbox" ${c.reasoning?'checked':''} style="margin-right:6px"><label for="llmReasoning" style="margin:0">支持推理</label>
-      </div>
-    </div>
-    <div class="form-group"><label>输入模式</label><div style="padding:8px 0">${inputModesChecked}</div></div>
-    <div class="flex gap-8 mt-16"><button class="btn btn-primary" onclick="saveLLM()">保存配置</button><button class="btn btn-success" onclick="testLLM()" onmouseenter="showTestCode(event,'global')" onmouseleave="hideTestCode()">测试连接</button></div>
-    <div id="testCodeTooltip" style="display:none;position:fixed;background:#1e1e1e;color:#d4d4d4;padding:12px;border-radius:8px;font-size:12px;font-family:Consolas,monospace;max-width:400px;z-index:9999;box-shadow:0 4px 12px rgba(0,0,0,0.3);border:1px solid #3c3c3c"><pre id="testCodeContent" style="margin:0;white-space:pre-wrap"></pre></div>
-</div>`;
+  let html='';
   // 用量统计
   html+=`<div class="card"><h3>用量统计</h3><div class="grid" style="grid-template-columns:repeat(4,1fr)">
     <div class="stat"><div class="label">调用次数</div><div class="value">${u.call_count||0}</div></div>
@@ -1179,11 +1142,12 @@ async function renderLLM(){
   if(!modelList.length){
     html+='<div class="empty">暂无自定义模型，点击上方按钮添加。</div>';
   }else{
-    html+='<table style="font-size:12px"><tr><th>ID</th><th>名称</th><th>Provider</th><th>模型</th><th>上下文</th><th>输入</th><th>推理</th><th>状态</th><th>操作</th></tr>';
+    html+='<table style="font-size:12px"><tr><th>ID</th><th>名称</th><th>Provider</th><th>模型</th><th>上下文</th><th>输入</th><th>推理</th><th>兜底</th><th>状态</th><th>操作</th></tr>';
     const providerColors={openai:'badge-green',anthropic:'badge-yellow',ollama:'badge-purple',zhipu:'badge-blue',custom:'badge-red',deepseek:'badge-blue',moonshot:'badge-purple',qwen:'badge-yellow',modelscope:'badge-purple'};
     for(const m of modelList){
       const badgeClass=providerColors[m.provider]||'badge-green';
       const inputModes=m.input_modes?m.input_modes.join(','):'text';
+      const isFallback=m.is_global_fallback!==false;
       html+=`<tr>
         <td><code>${escHtml(m.id)}</code></td>
         <td>${escHtml(m.name||'')}</td>
@@ -1192,9 +1156,10 @@ async function renderLLM(){
         <td>${(m.context_window||128000).toLocaleString()}</td>
         <td>${escHtml(inputModes)}</td>
         <td>${m.reasoning?'<span class="badge badge-purple">是</span>':'-'}</td>
+        <td><span style="cursor:help" title="作为保障系统运行的最终兜底模型">${isFallback?'<span class="badge badge-green">&#9745;</span>':'<span class="badge badge-red">&#9744;</span>'}</span></td>
         <td>${m.enabled?'<span class="badge badge-green">启用</span>':'<span class="badge badge-red">禁用</span>'}</td>
         <td><button class="btn btn-sm btn-ghost" onclick="showEditModelModal('${escHtml(m.id)}')">编辑</button>
-        <button class="btn btn-sm btn-success" onclick="testModel('${escHtml(m.id)}')" onmouseenter="showTestCode(event,'model','${escHtml(m.id)}','${escHtml(m.model)}','${escHtml(m.base_url)}','${escHtml(m.provider)}','${escHtml(m.api_type||'')}','${m.temperature||0.1}')" onmouseleave="hideTestCode()">测试</button>
+        <button class="btn btn-sm btn-success" onclick="testModel('${escHtml(m.id)}')">测试</button>
         <button class="btn btn-sm btn-danger" onclick="deleteModel('${escHtml(m.id)}','${escHtml(String(m.name||'').replace(/'/g,"\\'"))}')">删除</button></td></tr>`;
     }
     html+='</table>';
@@ -1202,70 +1167,12 @@ async function renderLLM(){
   html+='</div>';
   $('content').innerHTML=html;
 }
-async function saveLLM(){
-  const inputModes=Array.from(document.querySelectorAll('[id^="llmInputMode_"]:checked')).map(c=>c.value);
-  const apiTypeSelect=$('llmApiTypeSelect').value;
-  const apiType=apiTypeSelect==='custom'?($('llmApiTypeCustom').value||'custom'):apiTypeSelect;
-  const payload={
-    provider:$('llmProvider').value,
-    api_type:apiType,
-    model:$('llmModel').value,
-    base_url:$('llmUrl').value,
-    temperature:parseFloat($('llmTemp').value)||0.1,
-    max_tokens:parseInt($('llmMaxTokens').value)||4096,
-    context_window:parseInt($('llmContextWindow').value)||128000,
-    input_modes:inputModes,
-    reasoning:$('llmReasoning').checked
-  };
-  if($('llmKey').value)payload.api_key=$('llmKey').value;
-  const r=await api('/api/llm',{method:'PUT',body:JSON.stringify(payload)});
-  if(r.error){showToast(r.error,'danger');return}showToast('已保存','success');renderLLM();
-}
-async function testLLM(){
-  const btn=event.target;btn.textContent='测试中...';btn.disabled=true;
-  hideTestCode();
-  const provider=$('llmProvider').value,model=$('llmModel').value,base_url=$('llmUrl').value;
-  const temperature=parseFloat($('llmTemp').value)||0.1;
-  const apiTypeSelect=$('llmApiTypeSelect').value;
-  const api_type=apiTypeSelect==='custom'?($('llmApiTypeCustom').value||'custom'):apiTypeSelect;
-  const r=await api('/api/llm/test',{method:'POST',body:JSON.stringify({provider,api_type,model,base_url,temperature})});btn.textContent='测试连接';btn.disabled=false;
-  showToast(r.ok?'连接成功: '+r.response:'连接失败: '+r.error,r.ok?'success':'danger');
-}
-function showTestCode(e,type,modelId,modelName,baseUrl,provider,apiType,temperature){
-  const tooltip=document.getElementById('testCodeTooltip');
-  const content=document.getElementById('testCodeContent');
-  if(!tooltip||!content)return;
-  let code;
-  if(type==='global'){
-    const provider=$('llmProvider').value,model=$('llmModel').value,base_url=$('llmUrl').value;
-    const temp=parseFloat($('llmTemp').value)||0.1;
-    const apiTypeSelect=$('llmApiTypeSelect').value;
-    const api_type=apiTypeSelect==='custom'?($('llmApiTypeCustom').value||'custom'):apiTypeSelect;
-    code=`POST /api/llm/test\n\n{\n  "provider": "${provider}",\n  "api_type": "${api_type}",\n  "model": "${model}",\n  "base_url": "${base_url}",\n  "temperature": ${temp}\n}`;
-  }else{
-    const showApiType=apiType&&apiType!=='openai-completions';
-    const temp=temperature||0.1;
-
-    code=`POST /api/llm/test\n\n{\n  "provider": "${provider}"${showApiType?`,\n  "api_type": "${apiType}"`:''},\n  "model": "${modelName}",\n  "base_url": "${baseUrl}",\n  "temperature": ${temp}\n}`;
-  }
-  content.textContent=code;
-  tooltip.style.display='block';
-  const rect=tooltip.getBoundingClientRect();
-  let left=e.clientX+10,top=e.clientY+10;
-  if(left+400>window.innerWidth)left=e.clientX-420;
-  if(top+100>window.innerHeight)top=e.clientY-110;
-  tooltip.style.left=left+'px';tooltip.style.top=top+'px';
-}
-function hideTestCode(){
-  const tooltip=document.getElementById('testCodeTooltip');
-  if(tooltip)tooltip.style.display='none';
-}
+// saveLLM / testLLM / showTestCode / hideTestCode 已移除（全局模型配置卡片已删除）
 async function testModel(modelId){
   const model=allModelsCache.find(m=>m.id===modelId);
   if(!model){showToast('模型不存在','danger');return;}
-  hideTestCode();
   const btn=event.target;btn.textContent='测试中...';btn.disabled=true;
-  // 必须发送 api_key，否则后端会 fallback 到全局配置，导致测试不准
+  // 直接发送模型自身配置，使用独立的测试接口，不经过任何兜底/fallback逻辑
   const payload={model:model.model,base_url:model.base_url,provider:model.provider,temperature:model.temperature||0.1,api_key:model.api_key||''};
   if(model.api_type)payload.api_type=model.api_type;
   const r=await api('/api/llm/test',{method:'POST',body:JSON.stringify(payload)});
@@ -1286,6 +1193,7 @@ function showAddModelModal(editId){
   const inputModesArr=model&&model.input_modes?model.input_modes:['text'];
   const reasoning=model?model.reasoning:false;
   const maxTokens=model&&model.max_tokens?model.max_tokens:4096;
+  const isFallback=model?model.is_global_fallback!==false:true;
   const provOpts=['custom','openai','anthropic','ollama','zhipu','deepseek','moonshot','qwen','modelscope'].map(p=>`<option value="${p}" ${provider===p?'selected':''}>${p}</option>`).join('');
   const apiTypeOpts=['openai-completions','openai-chat','anthropic','ollama'].map(t=>`<option value="${t}" ${apiType===t?'selected':''}>${t}</option>`).join('');
   const isCustomApiType=apiType==='custom'||!['openai-completions','openai-chat','anthropic','ollama'].includes(apiType);
@@ -1323,6 +1231,9 @@ function showAddModelModal(editId){
       </div>
     </div>
     <div class="form-group"><label>输入模式</label><div style="padding:8px 0">${inputModesChecked}</div></div>
+    <div class="form-group" style="display:flex;align-items:center">
+      <input id="mlGlobalFallback" type="checkbox" ${isFallback?'checked':''} style="margin-right:6px"><label for="mlGlobalFallback" style="margin:0;cursor:help" title="作为保障系统运行的最终兜底模型">全局兜底</label>
+    </div>
     <div class="flex gap-8 mt-16"><button class="btn btn-primary" onclick="doSaveModel('${escHtml(editId)}')">${isEdit?'保存':'添加'}</button><button class="btn btn-ghost" onclick="closeModal()">取消</button></div>
   </div></div>`;
 }
@@ -1348,7 +1259,8 @@ async function doSaveModel(editId){
     max_tokens:parseInt($('mlMaxTokens').value)||4096,
     temperature:parseFloat($('mlTemp').value)||0.1,
     input_modes:inputModes,
-    reasoning:$('mlReasoning').checked
+    reasoning:$('mlReasoning').checked,
+    is_global_fallback:$('mlGlobalFallback').checked
   };
   if($('mlApiKey').value)payload.api_key=$('mlApiKey').value;
   let r;

package/skills/ASR/LICENSE.txt

@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2025 z-ai-web-dev-sdk Skills
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.